A distributed approach to weighted frequent Subgraph mining

In various application domains like Social Network Analysis, cheminformatics etc, graph based data representations are used to effectively represent the complex relationships among entities. Many graph mining approaches have been developed over the years; of which one of the most challenging tasks is frequent subgraph mining (FSM). Most of the existing FSM algorithms consider only the graph-based structure, the importance of entities involved or the strength of their relationships are not considered. Weighted graphs may be better suited for many cases such as social networks, mobile-communication networks, transportation networks etc. Thus there is a need for efficient weighted subgraph mining approaches. Only few methods have been developed for FSM in weighted graphs. In the present big data era, to handle the massive quantity of data generated from various applications, there is a huge demand for distributed computational approaches. Currently, there is no such distributed approach for weighted FSM. In this work, a distributed approach for weighted frequent subgraph mining using MapReduce framework (WFSM-MR) is proposed. The idea is that both data and the computation part, i.e. weighted subgraph mining, should be distributed for efficient processing. The method is used to discover the significant weighted frequent subgraphs from various edge-weighted graph datasets. Experimental results indicate the proposed approach is efficient, scalable, and can handle large graph datasets compared to the existing non distributed weighted FSM task.